The developers of the solution explained the replacement of silicon processors with living neurons by the cost-effectiveness of the latter. Following the biological data center in Melbourne, another one is planned to be opened in Singapore.
Biotechnology will save money
Operators of data processing centers (DPC) around the world complain about the increasing costs of their maintenance, the high cost of components (microchips) and the huge size of facilities, which continue to grow. Huge amounts of electricity are required to keep equipment fully operational, including bulky cooling systems.
The developers of the pioneering data center in Melbourne promise to solve at least two problems. Biocomputers require significantly less electricity to operate, and the amount of data they can process is much larger. According to the calculations announced by Cortical Labs, each module of the Data Center consumes about 30 watts. With the same amount of work, a modern microchip in the AI infrastructure consumes more than 1 kW, sometimes much more.
Principle of operation
Data in a biological data center is processed by neurons grown from human brain stem cells and connected to microelectrode arrays. These modules, called CL1, control the cells, signal transmission processes and coordinate calculations, the results of which are uploaded to cloud storage.
The Data Center in Melbourne consists of 120 CL1 modules. In Singapore, 20 modules are initially laid out with a later increase to 1,000. The data obtained in these modules can be accessed remotely in real time.
At the embryonic level
With all the advantages that the new generation data centers have, the expert community notes many unresolved issues. In particular, the mechanisms of running full-fledged computational algorithms on them and saving the results of neuron training remain unclear. The capabilities that biological Data Centers demonstrate now are mainly “training” to perform simple tasks like playing computer games. There is no question of a full-fledged replacement of traditional AI infrastructure.
Also, the maintenance of these data centers involves keeping neurons alive, which requires additional resources. Providing them with nutrient media is one of the top-priority tasks, the solution of which will allow developers to scale the technology.
The mechanism of retraining neurons after a task is completed remains unresolved. Optimal methods of their training, including for machine learning tasks, have not been studied in detail.
Biological Data Centers require completely different approaches in terms of training of maintenance specialists. Programming neurons implies a much more serious level of training. The knowledge that programmers have in the traditional sense is not enough.
Although the introduction of technology opens up a wide range of data processing possibilities and can bring AI infrastructure to a new stage of development, it should be recognized that it is at an embryonic level, and a lot of resources must be applied to set up all processes.
